The Models and Methods That Answer Wanamaker’s “Which Half” Question Today
Agencies know that consumers don’t put much thought into the messaging or advertisements that convince them to buy one product over another. Seeing thousands of ads per day, our audiences are continually exposed to different messaging at every turn, and at some point take the bait to buy. Brands and marketers, however, spend a lot of time and money building customer journeys and assigning credit to certain touch points so that they may accurately measure the ROI of campaigns.
As marketing pioneer John Wanamaker once said, “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” Enter, attribution models. The purpose of an attribution model is to assign credit to the touchpoint which led to a conversion. Traditionally, attribution models have been defined as either single-touch or multi-touch. As you may have guessed, a single-touch attribution model assigns 100% credit to one touchpoint (usually the very first or very last).
Today, brands are behaving much more like media companies, constantly promoting content across multiple platforms. If a Sephora campaign promoting foundation includes banner ads, Facebook posts, direct mail samples, emails with reviews, and in-feed Instagram placements, how can one definitively say which single touch point ultimately led to a purchase? How can we definitively determine how much influence a sponsored blog post had over a Youtube pre-roll video? Multi-touch attribution models (MTA)– which assign fractional credit to each touchpoint– are much more accurate than single-touch attribution models.
Multi-touch attribution models help answer CMO’s perennial questions: Which marketing executions are most successful, and where should I spend my dollars? As consumers’ attentions are stretched thinner and thinner, clients want to know that their messaging efforts can cut through the clutter. As such, advertisers are ever-tempted to employ an MTA model to confidently portray which media placements directly contributed to a conversion. But even as an updated approach to the antiquated single-touch attribution, are MTA models fool-proof? Before considering the implementation of a multi-touch attribution model, it’s vital to first understand the different approaches to attributing ads and recognize the potential benefits or downfalls.
MTA models are most commonly categorized by methodology. The two most common methodologies for multi-touch attribution are rules-based and algorithmic. For the sake of this blog post, we’ll consider the most common models within each methodology.
Methodology 1: Rules-Based–A subjective, human-defined model that attributes credit to multiple touchpoints along a customer journey.
In a U-Shape MTA model, the most credit is awarded to the first and last touchpoints, with the supporting touchpoints in the middle receiving less attribution.
A linear MTA model attributes equal weight to all touchpoints. Linear MTA models are often used by businesses with long-running campaigns in which constant messaging is important to break through to consumers. This model is also common for brands which are new to multi-touch attribution.
Time Decay Model
Time decay MTA models gradually attribute credit leading up to the last touch point. This MTA model assumes a “What have you done for me lately” kind of mentality.
Consider this journey for each rules-based MTA model: The seasons are changing, and you’re in the market for a new winter jacket. You begin by searching “Best Down Jackets” and read a sponsored review of Patagonia-brand jackets. Then, see social ads of Patagonia jackets, which leads you to their website catalog. From there, you watch YouTube reviews of different styles. After determining which style of jacket you like best, you browse Amazon for cheaper deals. Days later, you receive an email from Patagonia that directs you back to the website where you finally convert and purchase your new winter jacket.
- Under the U-Shape model, the first paid search result and the last email would both receive 40% attribution. All touch points in-between would equally split the remaining 20% attribution.
- Using the linear model, all touch points along your journey would be awarded equal credit.
- A time decay MTA model would gradually attribute more credit to each touchpoint leading up to the last, with the email receiving the most attribution.
Methodology 2: Algorithmic– An objective attribution model built upon machine-learning and data
Shapley Value Model
A derivation of Game Theory which fairly attributes credit to each touchpoint according to their “true” contribution. This MTA model more accurately outlines a customer journey while determining which specific touchpoints outperformed others.
Markov Chain Model
Each channel in the marketing mix is considered a “state” in a consumer journey. “For example, if a visitor comes to a site via Email, they become part of the ‘Email state’ which has an increased probability of conversion compared to someone who has not come into any marketing channel at all. Increases (or decreases) in conversion probability from this approach are then used as attribution weights to distribute conversions equitably” (Approach).
One primary difference between these two algorithmic-based models is that the Shapley Value model does not attribute sequence, whereas the Markov Chain model fundamentally accounts for the sequence in a consumer journey. This difference allows for users of the Markov Chain approach to more granularly measure a personalized journey; however, this also makes implementation more difficult. As such, the Shapley Value model is more broadly used and understood within the industry.
Think of rules-based and algorithmic MTA models like airlines: If you only fly a few times per year, chances are you’re likely to choose the airline that offers the most convenient time or the cheapest price. In this instance, an airline may leverage a rules-based MTA model, as there are fewer unknown factors affecting your consumer journey. However, if you travel weekly, you’re much more likely to make purchasing decisions based upon multiple considerations, such as your routine itinerary, the accommodations offered on flight, price, availability, rewards, and more. All of these things are attributed to which airline you choose to fly, and any adjustments may impact your booking decision.
Although attribution models– whether a simple rules-based model or a more sophisticated algorithmic model– offer brands insight into their touchpoints’ performance, no model is perfect. There are external factors to keep in mind, such as the influence of word of mouth as well as the balance of an offline/online relationship, e.g. how many people may stroll into a brick-and-mortar makeup store after seeing an online ad. And just as no specific MTA model will flawlessly attribute a campaign spend, nor will one client’s model cleanly transfer to another; broad objectives need to be considered. For example, a short-term promotional campaign may justify a time-decay model, while a robust global marketing campaign could necessitate a custom algorithmic approach. Even if your agency (or client) is not yet ready to onboard a multi-touch attribution model, you’ve now developed a foundational understanding of the variation of models, their purposes, and which type may best serve your clients and campaigns.